Pasar al contenido principal

HACIA UN CICLO DE VIDA AUTOMATIZADO DE DATOS CENTRADO EN LA IA

PID2023-152841OA-I00

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para Impulsar la Investigación Científico-Técnica y su Transferencia
Subprograma Subprograma Estatal de Generación de Conocimiento
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2023
Plan Estatal 2021-2023 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSITAT POLITECNICA DE CATALUNYA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 4
Encontrada(s) 1 página(s)

UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

CAPS: Cost-Aware ML Pipeline Selection

  • Kontaxakis, Antonios
  • Sacharidis, Dimitris
  • Abelló Gamazo, Alberto|||0000-0002-3223-2186
  • Nadal Francesch, Sergi|||0000-0002-8565-952X
  • Simitsis, Alkis
We present CAPS, a novel system for cost-aware pipeline selection in automated machine learning (AutoML). CAPS’ approach is orthogonal to the search strategies used by existing AutoML frameworks, enabling seamless integration with them. This integration benefits AutoML systems by reducing computational waste—time spent evaluating inefficient pipelines or exceeding time or memory constraints—through lightweight performance and cost estimation. To incorporate cost-awareness into AutoML, CAPS represents a set of pipelines as a directed hypergraph, estimates execution times for each function, and formulates the optimal subset selection as a constrained prize-collecting optimization problem. CAPS employs a simple yet effective greedy algorithm to approximate this optimal subset. Our evaluation shows that when CAPS prioritizes cost, it reduces waste up to 4×—saving up to 23 hours in 50 hours of pipeline evaluation. When CAPS balances the trade-off between performance and cost, it consistently achieves performance improvements of 5% to 15% over the state of the art within the same time budget across datasets. It also demonstrates that CAPS is compatible with complementary efficiency techniques such as early stopping and warm-starting, achieving speedups of up to 4.7×., This work has been partially supported by the H2020-MSCA-ITN2020 DEDS project (GA 955895), the EU Horizon Europe research and innovation programmes DataGEMS (GA 101188416), WiseFood (GA 101181895), and CREXDATA (GA 101092749), MLEvol (PID2024-156019OB-I00/AEI/10.13039/501100011033) , TALC (PID2023-152841/OA-I00/AEI/10.13039/501100011033) and by ERDF, EU. The computation infrastructure has been provided by the Consortium des Équipements de Calcul Intensif (CÉCI), funded by the Fonds de la Recherche Scientifique de Belgique (F.R.S.-FNRS) under Grant No. 2.5020.11 and by the Walloon Region., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

Supporting data discovery tasks at scale with FREYJA

  • Maynou Yelamos, Marc|||0009-0004-3908-4392
  • Nadal Francesch, Sergi|||0000-0002-8565-952X
Data lakes, massive repositories of heterogeneous data, were popularized by implementing a load-first model-later approach to data integration in contrast to the traditional model-first loadlater implemented in data warehouses. Yet, such settings create new challenges related to a) how to effectively discover relevant data and automatically find meaningful relationships; and b) how to do so efficiently at scale. In order to address them, we present Freyja, a data discovery system designed to navigate through vasts amounts of data and support users on finding joinable pairs of attributes. Freyja is based on the definition of a novel similarity measure, designed to automatically discovery semantically-relevant relationships. The adopted metric is tailored to data lake contexts, where datasets are denormalized and data from several domains are stored. In order to optimize and scale the computation of such metric, Freyja adopts a learning approach relying on profiles, succinct and lightweight representations of the underlying characteristics of the data attributes. Freyja leverages analytical databases to efficiently compute column profiles. In this demo, we demonstrate how Freyja, which is offered as a Python library, supports data augmentation tasks (i.e., augmenting training datasets with new features to improve the performance of a machine learning model)., This work has been partly supported by the Horizon Europe Programme under GA.101135513 (CyclOps) and the Spanish Ministerio de Ciencia e Innovación under project PID2023-152841OA-I00 / AEI/10.13039/501100011033 (TALC)., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

A framework for continual knowledge graph embedding

  • Pons Recasens, Gerard|||0000-0003-2225-3255
  • Bilalli, Besim|||0000-0002-0575-2389
  • Queralt Calafat, Anna|||0000-0003-2782-2955
Knowledge Graph Embeddings (KGEs) are widely used to support downstream tasks over Knowledge Graphs (KGs) by representing entities and relations in a low-dimensional vector space. However, real-world KGs are not static, as new facts are continuously added, requiring old embeddings to be updated and new embeddings to be learned without forgetting previously acquired information. Existing KGE libraries focus on the static KG scenario. Moreover, while recent works on Continual Knowledge Graph Embedding (CKGE) propose strategies to address this problem, they rely on standalone frameworks and only support simple KGE models, such as TransE, limiting their potential. We extended the KGE library PyKEEN to enable continual learning in KGs. Our extension (i) expands any unimodal KGE model in PyKEEN with new entities and relations, (ii) introduces an informed initialization method to improve the addition of new embeddings, (iii) incorporates CKGE-oriented evaluation metrics to assess both performance and knowledge forgetting and acquisition, and (iv) offers an intuitive training function with dashboard support, enabling easy experimentation and tracking of the evolution of embeddings across updates. A repository containing the scripts, an interactive notebook tutorial, and an accompanying video are available., This work is supported by the Horizon Europe Programme under GA.101135513 (CyclOps) and the Spanish Ministerio de Ciencia e Innovación under project PID2023-152841OA-I00/AEI/10.13039/50110 0011033 (TALC)., Peer Reviewed
Recercat. Dipósit de la Recerca de Catalunya
Publicación Contribución de congreso (ConferenceOutput)

Supporting data discovery tasks at scale with FREYJA

  • Maynou Yelamos, Marc
  • Nadal Francesch, Sergi
Data lakes, massive repositories of heterogeneous data, were popularized by implementing a load-first model-later approach to data integration in contrast to the traditional model-first loadlater implemented in data warehouses. Yet, such settings create new challenges related to a) how to effectively discover relevant data and automatically find meaningful relationships; and b) how to do so efficiently at scale. In order to address them, we present Freyja, a data discovery system designed to navigate through vasts amounts of data and support users on finding joinable pairs of attributes. Freyja is based on the definition of a novel similarity measure, designed to automatically discovery semantically-relevant relationships. The adopted metric is tailored to data lake contexts, where datasets are denormalized and data from several domains are stored. In order to optimize and scale the computation of such metric, Freyja adopts a learning approach relying on profiles, succinct and lightweight representations of the underlying characteristics of the data attributes. Freyja leverages analytical databases to efficiently compute column profiles. In this demo, we demonstrate how Freyja, which is offered as a Python library, supports data augmentation tasks (i.e., augmenting training datasets with new features to improve the performance of a machine learning model)., This work has been partly supported by the Horizon Europe Programme under GA.101135513 (CyclOps) and the Spanish Ministerio de Ciencia e Innovación under project PID2023-152841OA-I00 / AEI/10.13039/501100011033 (TALC)., Peer Reviewed, Postprint (published version)